CTGANN: Channel-Mixed and Temporal Gated Attention Neural Network for GNSS/INS Compensation by Predicting Pseudo-Velocity During GNSS Outages

Abstract

The global navigation satellite system (GNSS) and the inertial navigation system (INS) integrated navigation system provide continuous and high-accuracy positioning; however, positioning accuracy deteriorates during GNSS outages due to the accumulation of INS errors. To address this challenge, we propose an efficient and novel model named channel-mixed and temporal gated attention neural network (CTGANN) to compensate for INS errors during GNSS outages by predicting pseudo-velocity. Compared to pseudo-position compensation, pseudo-velocity prediction effectively mitigates the accumulation of model prediction errors, resulting in a more stable and reliable solution during extended GNSS outages. When GNSS signals are available, CTGANN learns the complex nonlinear relationship between INS and GNSS measurements. During GNSS unavailability, CTGANN generates pseudo-velocity GNSS measurements to compensate, thereby effectively suppressing the divergence of positioning errors. CTGANN leverages the time mixing layer to effectively capture the underlying temporal dependency patterns in the data, while the channel-mixing layer emphasizes critical features and reduces redundant information. The proposed model’s performance was evaluated through field tests, and results show that CTGANN significantly improves GNSS/INS positioning accuracy during GNSS outages, outperforming other models.